# facebookresearch/MobileLLM

MobileLLM Optimizing Sub-billion Parameter Language Models for On-Device Use Cases. In ICML 2024.

Repository: https://github.com/facebookresearch/MobileLLM
Canonical: https://ross.abutalabs.com/products/mobilellm
Language: Python
License: NOASSERTION
License Family: other
Last push: 2026-04-30T00:14:05+00:00

## Health v2 (maintenance only)
Score: 59/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 79, release rhythm 35, longevity 56
- inputs: {"age_days": 786, "days_push": 126, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1459, forks 89 (observed 2026-08-28T04:04:47.205125+00:00)

## What it is
Meta's training code for MobileLLM, a family of sub-billion parameter language models optimized for on-device use, published at ICML 2024. It implements design techniques like SwiGLU activation, deep-thin architectures, embedding sharing, and grouped-query attention, with follow-up MobileLLM-R1 reasoning models.

## Use cases
- train a small language model under 1B parameters
- build an LLM that runs on mobile or edge devices
- reproduce MobileLLM or MobileLLM-R1 training recipes
- pretrain a compact LLM for math and coding tasks
- research efficient LLM architecture design
- deploy a reasoning model with limited compute budget

## When to choose
- you need to train or fine-tune sub-billion parameter LLMs for on-device deployment
- you want SoTA small-model accuracy on commonsense reasoning, math, or coding benchmarks
- you want full training code, data prep, and recipes released by Meta research

## When to avoid
- you just want to run inference with an existing model rather than train one
- you need a production serving framework or mobile SDK rather than research training code
- you lack multi-GPU training infrastructure, as the code assumes 8-GPU nodes

## Facets
- artifact type: library
- maturity: active
- function: llm-training, machine-learning, deep-learning
- domain: large-language-models, machine-learning, mobile-development
- platform: python
- tags: on-device-llm, small-language-models, research-code, pytorch, icml-2024, model-training, edge-ai, gpu, linux

## Member repositories
- facebookresearch/MobileLLM (main) score 59

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:47.205125+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:35:27.421365+00:00, confidence not recorded.
  - readme: https://github.com/facebookresearch/MobileLLM (fetched 2026-08-28T04:04:47.205125+00:00, sha b1390badd356)
- Data as of 2026-08-30T08:39:29.467469+00:00.